1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB 2 /* Copyright (c) 2020 Mellanox Technologies. */ 3 4 #include <linux/refcount.h> 5 #include <linux/list.h> 6 #include <linux/rculist.h> 7 #include <linux/rtnetlink.h> 8 #include <linux/workqueue.h> 9 #include <linux/spinlock.h> 10 #include <linux/notifier.h> 11 #include <net/netevent.h> 12 #include <net/arp.h> 13 #include "neigh.h" 14 #include "tc.h" 15 #include "en_rep.h" 16 #include "fs_core.h" 17 #include "diag/en_rep_tracepoint.h" 18 19 static unsigned long mlx5e_rep_ipv6_interval(void) 20 { 21 if (IS_ENABLED(CONFIG_IPV6) && ipv6_stub->nd_tbl) 22 return NEIGH_VAR(&ipv6_stub->nd_tbl->parms, DELAY_PROBE_TIME); 23 24 return ~0UL; 25 } 26 27 static void mlx5e_rep_neigh_update_init_interval(struct mlx5e_rep_priv *rpriv) 28 { 29 unsigned long ipv4_interval = NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME); 30 unsigned long ipv6_interval = mlx5e_rep_ipv6_interval(); 31 struct net_device *netdev = rpriv->netdev; 32 struct mlx5e_priv *priv = netdev_priv(netdev); 33 34 rpriv->neigh_update.min_interval = min_t(unsigned long, ipv6_interval, ipv4_interval); 35 mlx5_fc_update_sampling_interval(priv->mdev, rpriv->neigh_update.min_interval); 36 } 37 38 void mlx5e_rep_queue_neigh_stats_work(struct mlx5e_priv *priv) 39 { 40 struct mlx5e_rep_priv *rpriv = priv->ppriv; 41 struct mlx5e_neigh_update_table *neigh_update = &rpriv->neigh_update; 42 43 mlx5_fc_queue_stats_work(priv->mdev, 44 &neigh_update->neigh_stats_work, 45 neigh_update->min_interval); 46 } 47 48 static bool mlx5e_rep_neigh_entry_hold(struct mlx5e_neigh_hash_entry *nhe) 49 { 50 return refcount_inc_not_zero(&nhe->refcnt); 51 } 52 53 static void mlx5e_rep_neigh_entry_remove(struct mlx5e_neigh_hash_entry *nhe); 54 55 void mlx5e_rep_neigh_entry_release(struct mlx5e_neigh_hash_entry *nhe) 56 { 57 if (refcount_dec_and_test(&nhe->refcnt)) { 58 mlx5e_rep_neigh_entry_remove(nhe); 59 kfree_rcu(nhe, rcu); 60 } 61 } 62 63 static struct mlx5e_neigh_hash_entry * 64 mlx5e_get_next_nhe(struct mlx5e_rep_priv *rpriv, 65 struct mlx5e_neigh_hash_entry *nhe) 66 { 67 struct mlx5e_neigh_hash_entry *next = NULL; 68 69 rcu_read_lock(); 70 71 for (next = nhe ? 72 list_next_or_null_rcu(&rpriv->neigh_update.neigh_list, 73 &nhe->neigh_list, 74 struct mlx5e_neigh_hash_entry, 75 neigh_list) : 76 list_first_or_null_rcu(&rpriv->neigh_update.neigh_list, 77 struct mlx5e_neigh_hash_entry, 78 neigh_list); 79 next; 80 next = list_next_or_null_rcu(&rpriv->neigh_update.neigh_list, 81 &next->neigh_list, 82 struct mlx5e_neigh_hash_entry, 83 neigh_list)) 84 if (mlx5e_rep_neigh_entry_hold(next)) 85 break; 86 87 rcu_read_unlock(); 88 89 if (nhe) 90 mlx5e_rep_neigh_entry_release(nhe); 91 92 return next; 93 } 94 95 static void mlx5e_rep_neigh_stats_work(struct work_struct *work) 96 { 97 struct mlx5e_rep_priv *rpriv = container_of(work, struct mlx5e_rep_priv, 98 neigh_update.neigh_stats_work.work); 99 struct net_device *netdev = rpriv->netdev; 100 struct mlx5e_priv *priv = netdev_priv(netdev); 101 struct mlx5e_neigh_hash_entry *nhe = NULL; 102 103 rtnl_lock(); 104 if (!list_empty(&rpriv->neigh_update.neigh_list)) 105 mlx5e_rep_queue_neigh_stats_work(priv); 106 107 while ((nhe = mlx5e_get_next_nhe(rpriv, nhe)) != NULL) 108 mlx5e_tc_update_neigh_used_value(nhe); 109 110 rtnl_unlock(); 111 } 112 113 struct neigh_update_work { 114 struct work_struct work; 115 struct neighbour *n; 116 struct mlx5e_neigh_hash_entry *nhe; 117 }; 118 119 static void mlx5e_release_neigh_update_work(struct neigh_update_work *update_work) 120 { 121 neigh_release(update_work->n); 122 mlx5e_rep_neigh_entry_release(update_work->nhe); 123 kfree(update_work); 124 } 125 126 static void mlx5e_rep_neigh_update(struct work_struct *work) 127 { 128 struct neigh_update_work *update_work = container_of(work, struct neigh_update_work, 129 work); 130 struct mlx5e_neigh_hash_entry *nhe = update_work->nhe; 131 struct neighbour *n = update_work->n; 132 bool neigh_connected, same_dev; 133 struct mlx5e_encap_entry *e; 134 unsigned char ha[ETH_ALEN]; 135 struct mlx5e_priv *priv; 136 u8 nud_state, dead; 137 138 rtnl_lock(); 139 140 /* If these parameters are changed after we release the lock, 141 * we'll receive another event letting us know about it. 142 * We use this lock to avoid inconsistency between the neigh validity 143 * and it's hw address. 144 */ 145 read_lock_bh(&n->lock); 146 memcpy(ha, n->ha, ETH_ALEN); 147 nud_state = n->nud_state; 148 dead = n->dead; 149 same_dev = READ_ONCE(nhe->neigh_dev) == n->dev; 150 read_unlock_bh(&n->lock); 151 152 neigh_connected = (nud_state & NUD_VALID) && !dead; 153 154 trace_mlx5e_rep_neigh_update(nhe, ha, neigh_connected); 155 156 if (!same_dev) 157 goto out; 158 159 list_for_each_entry(e, &nhe->encap_list, encap_list) { 160 if (!mlx5e_encap_take(e)) 161 continue; 162 163 priv = netdev_priv(e->out_dev); 164 mlx5e_rep_update_flows(priv, e, neigh_connected, ha); 165 mlx5e_encap_put(priv, e); 166 } 167 out: 168 rtnl_unlock(); 169 mlx5e_release_neigh_update_work(update_work); 170 } 171 172 static struct neigh_update_work *mlx5e_alloc_neigh_update_work(struct mlx5e_priv *priv, 173 struct neighbour *n) 174 { 175 struct neigh_update_work *update_work; 176 struct mlx5e_neigh_hash_entry *nhe; 177 struct mlx5e_neigh m_neigh = {}; 178 179 update_work = kzalloc(sizeof(*update_work), GFP_ATOMIC); 180 if (WARN_ON(!update_work)) 181 return NULL; 182 183 m_neigh.family = n->ops->family; 184 memcpy(&m_neigh.dst_ip, n->primary_key, n->tbl->key_len); 185 186 /* Obtain reference to nhe as last step in order not to release it in 187 * atomic context. 188 */ 189 rcu_read_lock(); 190 nhe = mlx5e_rep_neigh_entry_lookup(priv, &m_neigh); 191 rcu_read_unlock(); 192 if (!nhe) { 193 kfree(update_work); 194 return NULL; 195 } 196 197 INIT_WORK(&update_work->work, mlx5e_rep_neigh_update); 198 neigh_hold(n); 199 update_work->n = n; 200 update_work->nhe = nhe; 201 202 return update_work; 203 } 204 205 static int mlx5e_rep_netevent_event(struct notifier_block *nb, 206 unsigned long event, void *ptr) 207 { 208 struct mlx5e_rep_priv *rpriv = container_of(nb, struct mlx5e_rep_priv, 209 neigh_update.netevent_nb); 210 struct mlx5e_neigh_update_table *neigh_update = &rpriv->neigh_update; 211 struct net_device *netdev = rpriv->netdev; 212 struct mlx5e_priv *priv = netdev_priv(netdev); 213 struct mlx5e_neigh_hash_entry *nhe = NULL; 214 struct neigh_update_work *update_work; 215 struct neigh_parms *p; 216 struct neighbour *n; 217 bool found = false; 218 219 switch (event) { 220 case NETEVENT_NEIGH_UPDATE: 221 n = ptr; 222 #if IS_ENABLED(CONFIG_IPV6) 223 if (n->tbl != ipv6_stub->nd_tbl && n->tbl != &arp_tbl) 224 #else 225 if (n->tbl != &arp_tbl) 226 #endif 227 return NOTIFY_DONE; 228 229 update_work = mlx5e_alloc_neigh_update_work(priv, n); 230 if (!update_work) 231 return NOTIFY_DONE; 232 233 queue_work(priv->wq, &update_work->work); 234 break; 235 236 case NETEVENT_DELAY_PROBE_TIME_UPDATE: 237 p = ptr; 238 239 /* We check the device is present since we don't care about 240 * changes in the default table, we only care about changes 241 * done per device delay prob time parameter. 242 */ 243 #if IS_ENABLED(CONFIG_IPV6) 244 if (!p->dev || (p->tbl != ipv6_stub->nd_tbl && p->tbl != &arp_tbl)) 245 #else 246 if (!p->dev || p->tbl != &arp_tbl) 247 #endif 248 return NOTIFY_DONE; 249 250 rcu_read_lock(); 251 list_for_each_entry_rcu(nhe, &neigh_update->neigh_list, 252 neigh_list) { 253 if (p->dev == READ_ONCE(nhe->neigh_dev)) { 254 found = true; 255 break; 256 } 257 } 258 rcu_read_unlock(); 259 if (!found) 260 return NOTIFY_DONE; 261 262 neigh_update->min_interval = min_t(unsigned long, 263 NEIGH_VAR(p, DELAY_PROBE_TIME), 264 neigh_update->min_interval); 265 mlx5_fc_update_sampling_interval(priv->mdev, 266 neigh_update->min_interval); 267 break; 268 } 269 return NOTIFY_DONE; 270 } 271 272 static const struct rhashtable_params mlx5e_neigh_ht_params = { 273 .head_offset = offsetof(struct mlx5e_neigh_hash_entry, rhash_node), 274 .key_offset = offsetof(struct mlx5e_neigh_hash_entry, m_neigh), 275 .key_len = sizeof(struct mlx5e_neigh), 276 .automatic_shrinking = true, 277 }; 278 279 int mlx5e_rep_neigh_init(struct mlx5e_rep_priv *rpriv) 280 { 281 struct mlx5e_neigh_update_table *neigh_update = &rpriv->neigh_update; 282 int err; 283 284 err = rhashtable_init(&neigh_update->neigh_ht, &mlx5e_neigh_ht_params); 285 if (err) 286 goto out_err; 287 288 INIT_LIST_HEAD(&neigh_update->neigh_list); 289 mutex_init(&neigh_update->encap_lock); 290 INIT_DELAYED_WORK(&neigh_update->neigh_stats_work, 291 mlx5e_rep_neigh_stats_work); 292 mlx5e_rep_neigh_update_init_interval(rpriv); 293 294 neigh_update->netevent_nb.notifier_call = mlx5e_rep_netevent_event; 295 err = register_netevent_notifier(&neigh_update->netevent_nb); 296 if (err) 297 goto out_notifier; 298 return 0; 299 300 out_notifier: 301 neigh_update->netevent_nb.notifier_call = NULL; 302 rhashtable_destroy(&neigh_update->neigh_ht); 303 out_err: 304 netdev_warn(rpriv->netdev, 305 "Failed to initialize neighbours handling for vport %d\n", 306 rpriv->rep->vport); 307 return err; 308 } 309 310 void mlx5e_rep_neigh_cleanup(struct mlx5e_rep_priv *rpriv) 311 { 312 struct mlx5e_neigh_update_table *neigh_update = &rpriv->neigh_update; 313 struct mlx5e_priv *priv = netdev_priv(rpriv->netdev); 314 315 if (!rpriv->neigh_update.netevent_nb.notifier_call) 316 return; 317 318 unregister_netevent_notifier(&neigh_update->netevent_nb); 319 320 flush_workqueue(priv->wq); /* flush neigh update works */ 321 322 cancel_delayed_work_sync(&rpriv->neigh_update.neigh_stats_work); 323 324 mutex_destroy(&neigh_update->encap_lock); 325 rhashtable_destroy(&neigh_update->neigh_ht); 326 } 327 328 static int mlx5e_rep_neigh_entry_insert(struct mlx5e_priv *priv, 329 struct mlx5e_neigh_hash_entry *nhe) 330 { 331 struct mlx5e_rep_priv *rpriv = priv->ppriv; 332 int err; 333 334 err = rhashtable_insert_fast(&rpriv->neigh_update.neigh_ht, 335 &nhe->rhash_node, 336 mlx5e_neigh_ht_params); 337 if (err) 338 return err; 339 340 list_add_rcu(&nhe->neigh_list, &rpriv->neigh_update.neigh_list); 341 342 return err; 343 } 344 345 static void mlx5e_rep_neigh_entry_remove(struct mlx5e_neigh_hash_entry *nhe) 346 { 347 struct mlx5e_rep_priv *rpriv = nhe->priv->ppriv; 348 349 mutex_lock(&rpriv->neigh_update.encap_lock); 350 351 list_del_rcu(&nhe->neigh_list); 352 353 rhashtable_remove_fast(&rpriv->neigh_update.neigh_ht, 354 &nhe->rhash_node, 355 mlx5e_neigh_ht_params); 356 mutex_unlock(&rpriv->neigh_update.encap_lock); 357 } 358 359 /* This function must only be called under the representor's encap_lock or 360 * inside rcu read lock section. 361 */ 362 struct mlx5e_neigh_hash_entry * 363 mlx5e_rep_neigh_entry_lookup(struct mlx5e_priv *priv, 364 struct mlx5e_neigh *m_neigh) 365 { 366 struct mlx5e_rep_priv *rpriv = priv->ppriv; 367 struct mlx5e_neigh_update_table *neigh_update = &rpriv->neigh_update; 368 struct mlx5e_neigh_hash_entry *nhe; 369 370 nhe = rhashtable_lookup_fast(&neigh_update->neigh_ht, m_neigh, 371 mlx5e_neigh_ht_params); 372 return nhe && mlx5e_rep_neigh_entry_hold(nhe) ? nhe : NULL; 373 } 374 375 int mlx5e_rep_neigh_entry_create(struct mlx5e_priv *priv, 376 struct mlx5e_neigh *m_neigh, 377 struct net_device *neigh_dev, 378 struct mlx5e_neigh_hash_entry **nhe) 379 { 380 int err; 381 382 *nhe = kzalloc(sizeof(**nhe), GFP_KERNEL); 383 if (!*nhe) 384 return -ENOMEM; 385 386 (*nhe)->priv = priv; 387 memcpy(&(*nhe)->m_neigh, m_neigh, sizeof(*m_neigh)); 388 spin_lock_init(&(*nhe)->encap_list_lock); 389 INIT_LIST_HEAD(&(*nhe)->encap_list); 390 refcount_set(&(*nhe)->refcnt, 1); 391 WRITE_ONCE((*nhe)->neigh_dev, neigh_dev); 392 393 err = mlx5e_rep_neigh_entry_insert(priv, *nhe); 394 if (err) 395 goto out_free; 396 return 0; 397 398 out_free: 399 kfree(*nhe); 400 return err; 401 } 402